Research map / Probabilistic and scientific modeling / Variational and generative sampling
Learning Latent Permutations with Gumbel-Sinkhorn Networks
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use Sinkhorn Networks instead of the recurrent sorting baseline | Number sorting · U(0,1), N=15 | Prop. any wrong: 0.9 -> .0 | Helped |
| Use Sinkhorn Networks for larger sorting inputs | Number sorting · U(0,1), N=120 | Prop. any wrong: .01 | Helped |
| Use the proposed two-layer Sinkhorn network rather than DeepPermNet | ImageNet jigsaw · 3x3 | Kendall tau: .72 -> .73 | Helped |
| Use Gumbel-Sinkhorn rather than the rounding baseline | C. elegans neural identification · 10% known neurons, Hard | accuracy: .21 -> .26 | Helped |
| Use Gumbel-Sinkhorn rather than MCMC | C. elegans neural identification · 40% known neurons, Easy | accuracy: .85 -> .97 | Helped |
| Use an extreme low temperature, tau=0.01 (ablation) | Sinkhorn Network · Celeba jigsaw 5x5 | Kendall tau: .78 -> .62 | Hurt |
| Use an extreme high temperature, tau=100 (ablation) | Sinkhorn Network · Celeba jigsaw 5x5 | Kendall tau: .78 -> .47 | Hurt |
| Use one Sinkhorn iteration instead of the standard iterations (ablation) | Sinkhorn Network · Celeba jigsaw 5x5 | Kendall tau: .78 -> .72 | Hurt |
| Add regularization (ablation) | Gumbel-Sinkhorn · C. elegans, 40% known neurons, Hard | accuracy: .93 -> .96 | Helped |
| Add regularization (ablation) | Gumbel-Sinkhorn · C. elegans, 10% known neurons, Hard | accuracy: .23 -> .26 | Helped |
Closest papers by meaning
- Learning Permutation from Structure Without Supervision (ICML 2026 · similarity 0.87)
- Gradient Estimation with Stochastic Softmax Tricks (NeurIPS 2020 · similarity 0.85)
- Leveraging Recursive Gumbel-Max Trick for Approximate Inference in Combinatorial Spaces (NeurIPS 2021 · similarity 0.85)
- EquiTabPFN: A Target-Permutation Equivariant Prior Fitted Network (NeurIPS 2025 · similarity 0.84)
- Contrasting Multiple Representations with the Multi-Marginal Matching Gap (ICML 2024 · similarity 0.84)
- REVISITING MULTI-PERMUTATION EQUIVARIANCE THROUGH THE LENS OF IRREDUCIBLE REPRESENTATIONS (ICLR 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.